Table 1.
Variables tested in resource selection functions with hypothesized biological relevance and predicted effect on wolf and mountain lion habitat selection.
Fig 1.
Wolf and mountain lion study areas.
Polygons are 100% minimum convex polygons (MCPs) surrounding wolf (in blue) or mountain lion (in red) Global Positioning System- (GPS) or Very High Frequency- (VHF) collar locations during summer that were used for resource selection function (RSF) modelling and testing. Base map was reprinted from the U.S. Geological Survey National Map and is without copyright.
Fig 2.
Selection for road density in high versus low canopy cover by wolves on the Rocky Mountain Front.
Relative probability of selection for roads in high canopy cover (>35%, dashed line) versus low canopy cover (<35%, solid line) areas by wolves on the Rocky Mountain Front, MT, USA. This cutoff was calculated as the median value of canopy cover at available location.
Fig 3.
Relative odds of selection for roads by wolves.
Relative odds of selection for road density by pack from mixed effects logistic regression model of within-territory resource selection by wolves, plotted against mean road density in each pack’s territory. Dashed line at y = 1 indicates neutral selection.
Table 2.
Resource selection coefficients and confidence intervals (CIs) from top-ranked fixed effects and mixed-effects resource selection functions for wolves.
Fig 4.
Internal and external cross validation of wolf resource selection functions.
Rows indicate the study area in which data to develop each resource selection function (RSF) originated, and columns indicate the study area in which data to test each RSF originated. Frames ‘A’, ‘E’ and ‘I’ show results from internal k-folds cross validation, plotting area-adjusted frequency of use by wolves (y axis) per RSF bin (x axis) from fixed-effect logistic regression models developed with GPS collar data from wolves. Each line in these plots represents 1 out of 5 folds of data used to cross-validate RSF predictions. Mean Spearman correlations were calculated between each RSF bin rank and area-adjusted frequency of use in each bin. Frames ‘B’, ‘C’, ‘D’, ‘F’, ‘G’ and ‘H’ show results from external cross-validation with raw frequency of use (y axis) per equal-area binned RSF deciles (x axis). Spearman correlations were calculated between each RSF bin rank and frequency of use in each bin. *Note that area-adjusted frequency of use is on the y-axis for internal validation plots, whereas raw frequency of use is on the y-axis for external validation plots.
Fig 5.
Selection for topographic position in high versus low canopy cover by mountain lions in the Garnet Range.
Relative probability of selection for topographic position index (TPI) in high canopy cover (>54%, dashed line) versus low canopy cover (<54%, solid line) areas by mountain lions in the Garnet Range, MT, USA. This cutoff was calculated as the median value of canopy cover at available locations. A positive slope indicates selection for ridgelines and peaks, while a negative slope indicates selection for drainages.
Table 3.
Resource selection coefficients and confidence intervals (CIs) from top-ranked fixed effects resource selection functions for mountain lions.
Fig 6.
External cross validation of mountain lion resource selection functions.
Rows correspond with different resource selection functions (RSFs) developed for mountain lions using GPS collar data from the Garnet Range, MT, USA, and columns indicate the study area in which data to test each RSF originated. Frames ‘A’ and ‘D’ show results from internal k-folds cross validation, plotting area-adjusted frequency of use by mountain lions (y axis) per RSF bin (x axis) from fixed-effect logistic regression models developed with GPS collar data from mountain lions. Each line in these plots represents 1 out of 5 folds of data used to cross-validate RSF predictions. Mean Spearman correlations were calculated between each RSF bin rank and area-adjusted frequency of use in each bin. Frames ‘B’, ‘C’, ‘E’ and ‘F’ show results from external cross-validation with raw frequency of use (y axis) per equal-area binned RSF deciles (x axis). Spearman correlations were calculated between each RSF bin rank and frequency of use in each bin.
Fig 7.
Maps of mountain lion resource selection functions.
Predicted relative probability of selection from (A) ‘moderate ruggedness’ mountain lion resource selection function (RSF) in the Whitefish Range, MT, USA, and (B) ‘high ruggedness’ mountain lion RSF) on the Rocky Mountain Front, MT, USA. RSF models were tested on VHF telemetry data from Kunkel et al. [59] and Williams [66]. Base map reprinted from Montana hillshade map under a CC BY license, with permission from Montana State Library, original copyright 2002.
Fig 8.
Map of wolf resource selection function in the Whitefish Range.
Predicted relative probability of selection during summer from top-performing resource selection function (RSF) for wolves in the Whitefish Range, MT, USA. Reprinted from Montana hillshade map under a CC BY license, with permission from Montana State Library, original copyright 2002.